Experimental results of a new differential protection algorithm in a distribution transformer using singular values

This paper describes a method to identify inrush currents in a distribution transformer using the Singular Value Decomposition during transient conditions. The proposed method applies the SVD to a measurement matrix formed by the differential current per each phase normalized and filtered. The goal is to obtain the greatest singular value that brings the most variability highlighting the characteristic patterns in differential currents. A threshold based on Euclidean norm was established to differentiate between inrush and fault currents. If the magnitude of the largest singular value is above the threshold, an internal fault is determined, otherwise, an inrush current. The algorithm was implemented on MATLAB, and a broad array of experimental cases was made in two different three-phase transformers to validate its performance. The algorithm successfully differentiated inrush conditions from internal faults in all 100 cases.